[Ifeffit] Linear Combination Fitting in Athena-

[scalvin at slc.edu]

Hi Skya,

I'm not convinced there's anything strange about this. You're only
reporting the best fit in each case. Presumably the fit series using at
most 4 standards also generates the result of standard 1, 2, and 3 being a
possibility, and generates the same statistical measures of fit as in the
3 standard case.

The moral is that the BEST fit out of linear combinations is not all that
meaningful if there are multiple other fits of similar statistical
quality. Linear combination fitting can be used to narrow down the
possibilities, but as always shouldn't be used as a "black box."

(Of course, if one of the 4-standard fits has much better statistical
quality than any of the 3-standard fits, that might suggests that there
really ARE 4 components in the sample!)

--Scott Calvin
Sarah Lawrence College

> I'm doing linear combination fits using 10 standards total, and choosing
> to use at most 3 or 4 standards in the fit. The problem is that depending
> on which I choose, 3 or 4, the results vary significantly. Often one
> standard will show up in one fit using at most 3 standards and not in a
> fit using at most 4 standards. For example:
>
> Unknown A:
>
> Fit using at most 3 standards
> Standard 1= 0.213
> Standard 2= 0.374
> Standard 3= 0.413
>
> Fit using at most 4 standards
> Standard 1= 0.282
> Standard 2= 0.269
> Standard 4= 0.102
> Standard 5= 0.346
>
> Unknown B:
>
> Fit using at most 3 standards
> Standard 1= 0.278
> Standard 2= 0.745
>
> Fit using at most 4 standards
> Standard 2= 0.694
> Standard 3= 0.117
> Standard 4= 0.227
>
>
> Note that the same 10 standards are run in each case, its only the number
> I choose to use in the best fit that is varying. As you can see,
> hopefully, the results vary significantly. Any ideas as to why this might
> happen? Any advice you can give on attaining more consistent results would
> be greatly appreciated.
>
> Thank you!
>
> Skya Fawcett
> PhD Candidate
> Department of Geological Sciences and Geological Engineering
> Queen's University, Kingston, Canada
>